Your brokers will use AI. Your brokerage sets the rules it works under.
Broker AI gives each broker an AI assistant that knows how they run their business. It also gives the brokerage one place to decide which AI tools are allowed, what AI may and may not be used for, and what never gets typed into one.
Client-facing drafts are checked against the brokerage’s rules, and you see which brokers may not be checking. Every change is versioned. Every broker’s acceptance is on record.
Free during the pilot · Data stored and processed in Canada · No client data held
- Anthropic (Claude)Approved
- Devs.aiApproved
- OpenAI (ChatGPT)Pending review
- Any AI tool not approvedBlocked
- Consumer-facing credit decisions
- Adverse-action decisions
- Anything the brokerage compliance manual prohibits
- No client names, SINs, account numbers or credit details in any AI tool
- Business or enterprise AI accounts only, never personal or free ones
Every request
The AI tool is checked against the policy, not just once at sign-up.
Every version
Kept, with who changed it, when and why.
Every broker
Accepts the policy, dated and versioned. The brokerage can make it a condition of using AI.
“We can’t use free AI models.” Here’s how the brokerage enforces that.
Broker AI doesn’t decide your policy. It holds the policy your compliance officer writes, applies it to every broker, and keeps the record.
The brokerage chooses the AI tools
Mark each tool Approved, Pending review or Prohibited, and block anything not approved. A blocked tool is refused when it tries to connect, and on every request after that. A policy change takes effect on the broker’s next request.
Connections are verified, not self-reported
When Claude, ChatGPT or Devs.ai connects, Broker AI confirms which tool it is before granting access. A key a broker pastes in never counts as an approved tool. In a ChatGPT Business or Enterprise workspace, the admin adds the connection for everyone.
“No free AI”, two ways to enforce it
No connection shows which plan a broker is on — we tested ChatGPT’s free plan, and it connects. So either the policy requires business or enterprise accounts, which the major vendors say aren’t used to train their models by default, and each broker accepts it before their AI can use Broker AI; or the brokerage approves only AI accounts it provisions and pays for, and blocks everything else.
For brokerages that want to provision and control the AI themselves: models from the major AI providers on one platform, on seats the brokerage buys and manages, with workflows we can build there for brokerages. Devs.ai states it holds SOC 2 Type II and ISO 27001 and has zero-training agreements with its model providers. It is never required — Broker AI works the same with whichever AI tool the brokerage approves.
We earn a commission if you sign up for Devs.ai through us.
From policy to proof.
The brokerage sets the policy once
Approved tools, approved and restricted uses, data rules, and the brokerage’s compliance rules. Each save becomes a new version, recording who saved it and why.
Brokers join and accept it
Invite a whole office from a spreadsheet. Each broker reads the policy and accepts it. Editing a rule doesn’t ask them again; changing the AI policy does.
The AI tool is checked
On connection and on every request. Prohibited, pending and unapproved tools are refused, with a message telling the broker why.
The AI reads the policy first
Before it drafts anything, the broker’s AI is given the approved uses, the restricted uses, the data rules and the brokerage’s rules.
Drafts are checked, and gaps are shown
Against the brokerage’s rules and the broker’s own. The AI is told to hand anything touching a restricted use back to the broker instead of finishing it. Brokers who may not be checking are listed under Coverage gaps.
It all goes on record
Each check is logged with the time, the channel, what was caught, which AI tool asked and which policy version applied. So is every refused connection.
The demo is easy. Approval is the hard part.
An AI agent that replies to leads, chases documents and prepares forms looks great in a demo. Then someone asks who approved what it sent, what it was allowed to change, and how the brokerage would show an examiner. That is where most agent projects stop.
In IBM’s 2026 study of 2,000 technology executives, 77% said AI adoption is outpacing their governance, and 11% said they feel prepared for the AI agents coming next year. In McKinsey’s 2025 survey, no more than 10% in any business function said they are scaling agents.
Sources: IBM Tech Leader Study, via CIO Dive · McKinsey, The state of AI in 2025
What the brokerage needs before an agent touches a file, and what Broker AI holds
- A person approves every send
- Agents draft. The broker reads, approves and sends. Each draft is checked against the brokerage’s rules before the broker sees it.
- Required steps it can’t skip
- The brokerage sets the steps every file goes through. A file can’t move past a required step that isn’t recorded; moving it anyway takes the broker’s reason, and the supervisor sees it.
- Rules outside the agent
- Every agent reads the same rules from Broker AI, so changing a rule changes it for all of them, whichever AI model they run on.
- A record of all of it
- Every check, step, approval and override, with the time and the policy version. Never the client file.
We’re building agents for every stage of a mortgage file on this, from first lead to funded, and testing them with one broker now. They aren’t available to sign up for yet.
A record an examiner can open.
Each broker can download their record at any time, as spreadsheet files ready to hand to a principal broker or an examiner. Nothing in it is typed up for the occasion; it’s assembled from what Broker AI already logged.
Deliberately absent: drafts and client information. A check records which rules fired, not what was written or who it was for.
- rules-in-force.csvThe broker’s rules today, each with the document it came from
- brokerage-rules.csvThe brokerage’s rules today
- ai-use-policy.txtThe brokerage’s AI use policy, as written
- brokerage-policy-history.csvEvery version: date, reason, approved and prohibited tools
- ai-policy-acknowledgements.csvWhen this broker accepted the policy, and which version
- checks-log.csvEvery check that ran: when, which channel, what was caught, which AI tool, which policy version
What the brokerage gets, and what it deliberately doesn’t.
The brokerage admin sees
- An oversight report across every broker: who has accepted the AI policy, drafts checked and problems caught, which AI tools are in use, and every refused connection — downloadable as a spreadsheet
- Who has joined, each person’s role, and whether they’ve started setup
- The tools each broker uses (names only)
- The full version history of the rules and the AI policy
The brokerage never sees
- A broker’s clients, leads or pipeline
- A broker’s drafts
- A broker’s profile, voice or scoring
Brokerage rules are applied to a broker’s AI at the moment it works and are never copied into the broker’s profile. A broker who leaves takes none of the brokerage’s rules with them.
From one page to the whole office.
- The compliance officer sets the AI policy and the brokerage rules on one page.
- Invite brokers from a spreadsheet with the columns email, name, role, tools — up to 500 at a time. A bad row is reported without stopping the rest.
- Each broker opens their invite link, signs in without a password, and accepts the policy.
- Each broker builds their own profile. The core setup takes about 15 minutes.
- The brokerage’s rules apply from the moment a broker joins. Changes to the rules reach every broker’s next check.
Where your data lives, what we keep, and what we never hold.
The facts a compliance officer or IT reviewer needs before approving Broker AI. Where something isn’t in place yet, this says so.
Optional Microsoft or Google sign-in, and Meta or Google advertising on the Growth plan, use those companies’ own services, and only when a person chooses them.
Stored
- How each person works: their profile, voice and compliance rules, and every version of them
- The firm’s rules and AI use policy, and every version
- A log of each draft check: when, which channel, which rules fired, which AI tool, which policy version
- Membership, invites, policy acceptances and refused AI connections; sign-in email addresses
Never held
- Drafts — each is checked in memory and discarded
- Client names, contact details, income or credit information
- Passwords — people sign in with a one-time email code, or with Microsoft or Google
- Anything sent to an AI company by Broker AI itself — the person’s own approved AI tool does the work
How it’s protected
- Encryption. HTTPS for everything in transit; the database is encrypted at rest. Ad-platform tokens are additionally encrypted (AES-256-GCM), and access keys and sign-in tokens are stored only as one-way hashes.
- Access to records. Row-level security on every table: each person reads only their own records, and a firm admin only their firm’s.
- Production access. One person, Broker AI’s co-founder, with two-factor sign-in on every production account.
- Backups. The provider’s daily backups, plus our own nightly copy, encrypted before it leaves the backup job and stored in Canada under a 30-day deletion lock. Each month the newest copy is restored into a separate database and checked.
Your data, your control
- Export, any time
- A firm admin downloads everything the firm owns in one step: rules and AI policy with every version, membership history, invites, acceptances, refused connections and check logs. Where the firm is set up as the employer, each member’s own data is included too, and members are told so when they join.
- Deletion
- On request, once the firm has its export. Removed from the live system straight away; it ages out of the backups within 90 days.
- If something goes wrong
- We notify affected firms as soon as we become aware of a breach.
- Contracts
- We will sign the firm’s data processing agreement.
- Not in place yet
- No SOC 2 report or independent penetration test yet. Cyber and errors-and-omissions cover will be in place before the first paid firm engagement.
Free during the pilot. Then a flat price per broker.
Business AI profile plus governance: drafts checked against the broker’s rules and the brokerage’s, the AI use policy, and the compliance record.
Everything in Broker AI, plus Meta and Google advertising checked against credit-ad rules and a spending ceiling. The broker approves every change.
- Brokerages with 50 or more brokers get group pricing.
- Annual billing: two months free.
- The brokerage decides who pays: the brokerage for everyone, the brokerage for the base plan with brokers adding Growth themselves, or each broker on their own.
- Invoiced to the brokerage with tax added, or paid by card.
Prices cover Broker AI itself. The AI tool’s own plan and any ad spend are separate. Nothing is charged during the pilot, and you’ll get notice before it ends.
Broker AI records and applies the brokerage’s own AI policy and rules. It is not legal advice, and it does not certify compliance with any regulation. The brokerage decides its policy, and the licensed broker remains responsible for what goes out under their name.
It controls what connects to Broker AI. It can’t stop someone copying text into an unapproved tool by hand; the written policy and each broker’s recorded acceptance cover that.
Broker AI is a young company. It’s hosted on established providers and built to hold as little as possible, but it doesn’t yet have the certifications a larger vendor would. If anything here matters to your approval, ask and we’ll show you.
Set the rules once. Every broker’s AI works inside them.
We’ll walk your compliance officer through the policy, the oversight report and the record, and send the two one-page summaries — AI governance, and security and data — for your files.
russ@disruptionio.com · 778-658-6990